隨著腦科學研究發展，腦電波的分析讓人們揭開大腦神秘的面紗，腦機介面建構了一條腦波與其他機器之間溝通的管道，讓複雜的腦波經過處理能讓電腦順利分析，甚至是直接用腦波控制電腦。在許多研究中，腦電波裡特定的頻寬與大腦活動和動作表現有關聯，因此本研究希望藉由整合腦波訊號處理與自製模擬駕駛系統，提供足夠且標記的腦波訓練資料，並以前一秒的腦波來預測0.25秒後的方向盤角度，以此發展貼近日常駕駛習慣的腦波駕駛系統。本研究分為兩個部分，第一部分為預訓練實際運動模型，以PC方向盤作為標記工具，在虛擬道路中記錄不同時刻方向盤的角度作為腦波的分類標記，並結合連續時間的方向盤角度與腦波資訊，藉此來訓練連續時間實際運動模型；第二部分為使用多次遷移學習方式，來訓練想像運動腦波駕駛系統，以期望達到相同效果。兩個部分的實驗中皆將乾式腦波電極設置在10-20 EEG System之C3、Cz、C4、Fz、Fp1、Fp2、P3、P4的位置，標記方式為動作瞬間作為基準點，以此基準點向前取一秒的資料作為分析腦波的區間，透過小波變換(wavelet transform)的方式取出特定的腦波頻段，再將八個通道所取出的頻率與時間關係做疊加，疊加後的二維資料，最後送入長短期記憶神經網路(Long Short-Term Memory, LSTM)分析使用者腦波變化。目前第一部分實際運動三分類已經達到受測者們平均80.36% ± 1.92%的辨識率，並在第二部分想像運動腦波駕駛中達到57.38%± 1.48%準確率。;With the development of EEG science research, the advanced techniques for brain wave analysis enable people to probe the profound neurocircuitry inside human brain. One promising technique is the use of brain waves to control peripheral machines through user’s intentions, which is called brain copmuter interface (BCI). The major challenge in designing a BCI is its signal processing to extract useful inteion informamtion so that the brain wave can be used to control external devices smoothly. According to the researches in the past, some specific frequency bands in the human brain are related to user’s limb movements. In this study, we intend to develop a brainwave-controlled driving system in vitrual reality (VR) driving environment. This study is divided into two parts. The first part is the pre-trained actual motion model. By gathering sufficient data for neural network training, the gathered continuous EEG were integrated with labeled steering information in driving environment. These two datas are used to train the continuous-time actual motion model. The second part is the use of multiple transfer learning methods to train the imaginary brain wave driving system in order to achieve the same effect. In this study, the EEG data were recorded from C3, Cz, C4, Fz, Fp1, Fp2, P3, P4 electrode positions based on the international 10-20 momntage system.The one-second EEG data preceding the momvement point was used as input data for BCI control. The one-second eight-channel EEG data were transformed into temprospectral domamin using wavelet transform and then used as input data for Long Short-Term Memory (LSTM) network to identify user’s momvement intention. The pre-trained actual motion model has achieved an accuracy rate of 80.36%± 1.92% and the brain wave driving system has achieved 57.38%± 1.48%.
